Detecting Toe-Off Events Utilizing a Vision-Based Method.

Area of Science:

• Biomechanics
• Computer Vision
• Machine Learning

Background:

• Gait event detection is crucial for analyzing human movement.
• Current methods often rely on wearable sensors, posing limitations like user cooperation and power constraints.
• There is a need for non-invasive, user-independent gait analysis techniques.

Purpose of the Study:

• To develop a novel algorithm for accurate gait event detection, specifically toe-off events.
• To enable gait analysis using a single 2D vision camera without requiring participant cooperation.
• To overcome the limitations of sensor-based gait detection methods.

Main Methods:

• Proposed a novel feature representation: consecutive silhouettes difference maps (CSD-maps).
• CSD-maps encode sequential pedestrian silhouettes from video frames to capture gait patterns.
• Utilized a convolutional neural network (CNN) for feature dimension reduction and toe-off event classification.

Main Results:

• The proposed CSD-maps effectively represent gait patterns for event detection.
• The CNN model successfully classified toe-off events based on CSD-map features.
• Experiments on a public database confirmed the method's good detection accuracy.

Conclusions:

• The novel CSD-map feature combined with CNN offers an accurate and non-invasive approach to gait event detection.
• This vision-based method removes the need for wearable sensors, enhancing user convenience and applicability.
• The algorithm shows promise for advancing gait analysis in various applications.